Proving Linear Mode Connectivity of Neural Networks via Optimal Transport

Damien Ferbach, Baptiste Goujaud, Gauthier Gidel, Aymeric Dieuleveut
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:3853-3861, 2024.

Abstract

The energy landscape of high-dimensional non-convex optimization problems is crucial to understanding the effectiveness of modern deep neural network architectures. Recent works have experimentally shown that two different solutions found after two runs of a stochastic training are often connected by very simple continuous paths (e.g., linear) modulo a permutation of the weights. In this paper, we provide a framework theoretically explaining this empirical observation. Based on convergence rates in Wasserstein distance of empirical measures, we show that, with high probability, two wide enough two-layer neural networks trained with stochastic gradient descent are linearly connected. Additionally, we express upper and lower bounds on the width of each layer of two deep neural networks with independent neuron weights to be linearly connected. Finally, we empirically demonstrate the validity of our approach by showing how the dimension of the support of the weight distribution of neurons, which dictates Wasserstein convergence rates is correlated with linear mode connectivity.

Cite this Paper

BibTeX
@InProceedings{pmlr-v238-ferbach24a, title = { Proving Linear Mode Connectivity of Neural Networks via Optimal Transport }, author = {Ferbach, Damien and Goujaud, Baptiste and Gidel, Gauthier and Dieuleveut, Aymeric}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {3853--3861}, year = {2024}, editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen}, volume = {238}, series = {Proceedings of Machine Learning Research}, month = {02--04 May}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v238/ferbach24a/ferbach24a.pdf}, url = {https://proceedings.mlr.press/v238/ferbach24a.html}, abstract = { The energy landscape of high-dimensional non-convex optimization problems is crucial to understanding the effectiveness of modern deep neural network architectures. Recent works have experimentally shown that two different solutions found after two runs of a stochastic training are often connected by very simple continuous paths (e.g., linear) modulo a permutation of the weights. In this paper, we provide a framework theoretically explaining this empirical observation. Based on convergence rates in Wasserstein distance of empirical measures, we show that, with high probability, two wide enough two-layer neural networks trained with stochastic gradient descent are linearly connected. Additionally, we express upper and lower bounds on the width of each layer of two deep neural networks with independent neuron weights to be linearly connected. Finally, we empirically demonstrate the validity of our approach by showing how the dimension of the support of the weight distribution of neurons, which dictates Wasserstein convergence rates is correlated with linear mode connectivity. } }
Endnote
%0 Conference Paper %T Proving Linear Mode Connectivity of Neural Networks via Optimal Transport %A Damien Ferbach %A Baptiste Goujaud %A Gauthier Gidel %A Aymeric Dieuleveut %B Proceedings of The 27th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2024 %E Sanjoy Dasgupta %E Stephan Mandt %E Yingzhen Li %F pmlr-v238-ferbach24a %I PMLR %P 3853--3861 %U https://proceedings.mlr.press/v238/ferbach24a.html %V 238 %X The energy landscape of high-dimensional non-convex optimization problems is crucial to understanding the effectiveness of modern deep neural network architectures. Recent works have experimentally shown that two different solutions found after two runs of a stochastic training are often connected by very simple continuous paths (e.g., linear) modulo a permutation of the weights. In this paper, we provide a framework theoretically explaining this empirical observation. Based on convergence rates in Wasserstein distance of empirical measures, we show that, with high probability, two wide enough two-layer neural networks trained with stochastic gradient descent are linearly connected. Additionally, we express upper and lower bounds on the width of each layer of two deep neural networks with independent neuron weights to be linearly connected. Finally, we empirically demonstrate the validity of our approach by showing how the dimension of the support of the weight distribution of neurons, which dictates Wasserstein convergence rates is correlated with linear mode connectivity.
APA
Ferbach, D., Goujaud, B., Gidel, G. & Dieuleveut, A.. (2024). Proving Linear Mode Connectivity of Neural Networks via Optimal Transport . Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:3853-3861 Available from https://proceedings.mlr.press/v238/ferbach24a.html.

Related Material